Open philnik777 opened 1 month ago
@llvm/issue-subscribers-clang-frontend
Author: Nikolas Klauser (philnik777)
Seems like there should be one for masked load too. For masked load, you might want a _z (zero) and a _m (merge) version or just have a merge version which will be optimized when the merge vector is zero
.
LLVM has an overloaded
llvm.masked.store
intrinsic, but there is no generic way to do a masked store from C/C++ code. AFAICT it's only exposed via target-specific intrinsics. It would be great if there could be a__builtin_masked_store
or something like that to make it easier to write generic code.